Pytorch3d: Create a toy example for SfMPerspectiveCameras

[IDSPAGET]

🚀 Feature

Add a small toy example of rendering with SfMPerspectiveCameras .

Motivation

A lot of users seem to have issues with making SfMPerspectiveCameras work (#154,#148,#146, #18, #184 ...).
Maybe a toy example would bring more non-computer graphics users to use the API.

Pitch

A simple script that load a mesh, an image and a calibration in a simple intrinsic/extrinsic format and render the corresponding RGB frame.
Scannet could be used for instance.

[gkioxari]

Hi @IDSPAGET
Indeed, users seem confused regarding cameras defined in screen or ndc space. We are in the process of fixing it and we will make sure to add examples.

[BostonLobster]

@gkioxari @IDSPAGET Hi, may I ask what is the current progress of this toy example. I try to render a RGB image using point cloud from ScanNet, but failed with an empty frame. Hope you could release the example ASAP🙏

[gkioxari]

Hi @BostonLobster @IDSPAGET
We have updated the docs for cameras here

https://github.com/facebookresearch/pytorch3d/blob/master/docs/notes/cameras.md

and you can read more in our codebase regarding the details of the camera parameters. Let us know if that is useful!

[nikhilaravi]

We will close this issue. Feel free to reopen it if you have more questions.

[BostonLobster]

@IDSPAGET Have you successfully rendered a ScanNet scene? I still stuck on weird or empty result for rendering a ScanNet point cloud.